When caution becomes a cost
A study that recently appeared in Harvard Business School Working Knowledge touches on a seemingly niche but in fact very serious topic. Rembrand Koning and colleagues from UC Berkeley and Stanford analysed 18 independent studies involving more than 140,000 students and workers across several countries – from the United States to Morocco. The conclusion is clear: women use generative AI on average 25% less often than men.
The number itself is not surprising. What is more puzzling, however, is what it does not explain. In one experiment, 17,000 Kenyan entrepreneurs were given the exact same access to ChatGPT and the same instructions. Despite this, women were still 13% less likely to use the tool. Access alone is clearly not enough.
Why? Koning suggests that women’s behaviour stems not from unfamiliarity with the technology, but from something else: a fear that using AI will be perceived as a kind of cheating or an admission of one’s own lack of competence. “Women face greater penalties in being judged as not having expertise in different fields,” the researcher says. In practice, this means that the same gesture – typing a question into the tool – is read differently depending on who is performing it. A man reaches for an effective tool. A woman – “cheats”.
It is difficult not to dwell on this asymmetry for a while. After all, this is not about fear of technology, or even about a lack of curiosity. It is about a restraint that comes from reading the social context – a context which, in many organisations, still punishes women more severely for the same thing that, in men, passes for competence. This is not an individual choice of caution. It is a rational calculation in an environment that is still not neutral.
Koning warns of three consequences. First – the competence gap may deepen, and with it the disparity in salaries and promotions. Second – companies lose potential productivity gains. Third, and perhaps least obviously, large language models learn mainly from data and interactions provided by men, which may entrench biases within the technology itself. Women’s reticence is therefore not without trace in the very tools they themselves are reluctant to use.
What follows from this? The author notes that where leaders openly say “we want everyone to try; some things will work, some will not — and that’s okay”, the adoption gaps are the smallest. He draws on Amy Edmondson’s concept of psychological safety – an environment in which one can make mistakes, experiment and ask for help without losing credibility.
Perhaps this is where the real answer lies. Not so much in how we teach people to use AI – but in whether we create conditions in which they are allowed to use it without worrying about how they will be read.
This post is part of the project “People and Algorithms in Organisations: Competences to Work in the Digital Environment” (DIGIT_People and algorithms), funded by the NAWA – Narodowa Agencja Wymiany Akademickiej).